Virtual Network Mapping Algorithm Optimization And Implementation Research

The rapid development of Internet technology has had a profound impact on social development,but the traditional network architecture can hardly cope with the challenges brought by the increasing scale of the network,which has become a bottleneck restricting the development of the digital economy.Network virtualization technology realizes dynamic mapping and optimal allocation of network resources by building virtual networks,effectively releasing the network potential and is a key way to break through the bottleneck of network development.Among them,virtual network embedding technology can efficiently map physical network resources to virtual network requests dynamically,which is the core of realizing network virtualization.With the diversification of user requirements and the complexity of network applications,the scale of virtual network embedding problems is increasing,and traditional static algorithms and existing machine learning algorithms are hardly capable of handling them.Therefore,in order to solve these problems,it is necessary to optimize the virtual network mapping algorithm based on the original work so that it can be better applied to large-scale VNE problems.Moreover,the algorithm performance metrics in these aspects,such as cost/benefit,and virtual network request acceptance rate,are used to measure the excellence of the VNE algorithm.The main research of this thesis is as follows.First,benefiting from the attention mechanism and reinforcement learning theory,this paper proposes GR-VNE,a virtual network mapping algorithm based on the attention mechanism and reinforcement learning strategy,which makes the model pay more attention to the physical nodes that may be selected and reduce the reliance on external information by introducing the attention mechanism.Through the analysis of simulation experimental results,the GR-VNE algorithm has good flexibility,and the algorithm outperforms other comparative algorithms in terms of virtual network request acceptance rate.Second,some problems on VNE algorithms based on machine learning and deep learning models are addressed,such as manual manual feature extraction affects the training process,the demand of resources for future virtual network requests is not considered,and how to reduce the fragmentation of resources.Therefore,this paper proposes a virtual network mapping algorithm based on the generative adversarial network model for optimization.The algorithm first extracts the potential attribute features from the physical network by using the generative adversarial network model,and uses a spectral clustering algorithm to find the embedded regions with energy-saving potential for the virtual network.After the experiments in Chapter 4,it is demonstrated that the GAN-VNE method proposed in this paper is not only effective in reducing the fragmentation of the physical network,but also significantly improves the acceptance rate and user experience,long-term benefit/cost ratio of virtual network requests.